prs-eth

prs-eth / PaGeR

Public

PaGeR — Unified Panoramic Geometry Estimation via Multi-View Foundation Models

48
1
94% credibility
Found Jun 02, 2026 at 48 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

PaGeR is a research tool that takes a single 360° panoramic image and automatically figures out the 3D geometry of the scene. It predicts how far away everything is (depth), which direction surfaces are facing (normals), and identifies the sky. Users can upload panoramas through a web interface and see results as colorful depth maps, surface direction maps, or interactive 3D point clouds. The tool can also process batches of images for evaluation against ground truth data.

How It Works

1
📷 You have a 360° photo

You've captured a panoramic image and want to understand its 3D structure.

2
🔍 You upload your panorama

You drag and drop your 360° image into the web interface and watch it load.

3
✨ The magic happens automatically

The AI studies your image and figures out how far away every surface is, plus which way each wall and object is facing.

4
🗺️ You see depth and surface maps

The results appear as colorful maps showing distance (warm colors = closer, cool = farther) and surface directions.

5
Choose how to explore
🗺️
Keep exploring the maps

Zoom in and out of the depth and normals maps to examine details.

🌐
View as a 3D cloud

Switch to a 3D point cloud where you can rotate and explore the scene from any angle.

🎉 You extracted 3D geometry from your photo

Your panorama is now a rich 3D understanding with depth measurements and surface directions you can use for any project.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 48 to 48 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is PaGeR?

PaGeR is a Python-based panoramic geometry estimation tool that takes a single 360-degree equirectangular image and produces depth maps, surface normals, and sky segmentation in one forward pass. Built on a Vision Transformer foundation model with a ViT-Giant backbone, it handles both indoor and outdoor scenes automatically by routing through specialized scale prediction heads. The model runs on PyTorch with CUDA acceleration and ships with a Gradio demo for quick experimentation.

Why is it gaining traction?

Unlike traditional multi-view stereo pipelines, PaGeR works from a single panorama with no camera pose estimation required. The unified approach consolidating depth, normals, and sky prediction into a single model eliminates the need to chain multiple specialized tools together. Its ability to output metric depth values directly in meters rather than just relative values addresses a common pain point in 3D reconstruction workflows. A free Gradio demo hosted on Hugging Face lowers the barrier to experimentation.

Who should use this?

This is particularly useful for 3D artists and game developers building environments from panoramic photography, robotics researchers needing 360-degree spatial awareness, and developers working with street-level or indoor mapping datasets. Heritage digitization teams and VR content creators working with panoramic captures will find the streamlined pipeline valuable. Academic researchers exploring panoramic depth estimation can benefit from the published datasets and evaluation code.

Verdict

PaGeR shows real technical innovation but carries a 0.949999988079071% credibility score reflecting its early stage: 48 stars, fresh release, and minimal community feedback. The non-commercial license on model weights is a significant constraint for commercial applications. Worth evaluating via the Hugging Face demo, but validate thoroughly before production deployment.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.